Accelerating Approximate Pattern Matching with Processing-In-Memory (PIM) and Single-Instruction Multiple-Data (SIMD) Programming Damla Senol Cali1 Zülal Bingöl2, Jeremie S. Kim1,3, Rachata Ausavarungnirun1, Saugata Ghose1, Can Alkan2 and Onur Mutlu3,1 1 2 3

Poster SEQ-15 CPU Memory Problem: Bitap algorithm can perform string matching with fast and simple bitwise operations. Due to the von Neumann architecture, memory bus between the CPU and the memory is the bottleneck of the high-throughput parallel bitap computations. Goals: Overcoming memory bus bottleneck of approximate string matching by performing processing-in-memory to exploit the high internal bandwidth available inside new and emerging memory technologies, and Using SIMD programming with Xeon Phi to take advantage of the high amount of bit parallelism available in the bitap algorithm. CPU Memory Low capacity memory bus The operations used during bitap can be performed in parallel, but high-throughput parallel bitap computation requires a large amount of memory bandwidth that is currently unavailable to the processor.  Memory bus is the bottleneck. This is a problem for all approximate string matching problems but bitap is PIM- and SIMD-friendly since it is entirely based on bitvectors and elementary bitwise operations. PIM -> perform bitwise operations near memory by bypassing the memory bus and with the help of many DRAM rows throughput is pretty high. SIMD -> we use its local high bandwidth memory/cache. Goal is again bypassing the memory bus in the offload mode.